stable-diffusion-webui-docker
stable-diffusion-webui-docker copied to clipboard
AbdBarho
Add support and enable deepdanbooru setting
You ought have created an issue first as per Readme
That being said, I wanted to add this feature a while ago, but I was very hesitant on making the image 4GB bigger and 2 times slower to build.
Would it make sense to have this as a build time flag?
Sure, let me look into it. Honestly I just did the PR by accident in the midst of like 90 other things and went with it, sorry about that.
I'm not as well versed with profiles, so I don't know if this can be done in a more elegant manner. I wasn't able to quickly get it to inherit the build context from the base profile, so it had to be redefined.
Another option would be to not have a second profile and do this, not sure what your feelings are on this.
@pirahtays I like the second option, but the user should not comment in the docker-compse.yml file, instead use a docker-compose.override.yml:
services:
auto:
build:
args:
- DEEPDANBOORU=1
environment:
- CLI_ARGS=--allow-code --medvram --xformers --deepdanbooru
Also, freeze the versions of tensorflow and co in the install command, and move the entire install block before the repo / deps update command, because that one is updated very often, and I don't want users to re-install tensorflow on every update.
The changes work fine, but I noticed that on 1 of my 2 SD machines CUDA functions fail after running an "interrogate deepbooru" step. Looking into if there are any missing packages / configuration for this.
I resolved this, just need to check in the fix. it appears that using tensorflow and not tensorflow-cpu causes all sorts of weirdness with cuda conflicts at times.
@pirahtays thank you, I created a new PR #194 with small tweaks after testing on my machine, I will close this one. You will be credited.
@pirahtays thank you, I created a new PR #194 with small tweaks after testing on my machine, I will close this one. You will be credited.
👍